Modern data processing systems typically comprise a host computer, consisting of an arithmetic and logic unit and a main memory unit for containment of data and instructions presently being processed, and long-term storage means for storage of data and processing instructions at other times. The long-term storage means is typically connected to the host computer by means of a channel. When the host desires a particular data set or record, it issues a command over the channel to the long-term storage means, which then reads the data, from whatever medium it is stored upon, e.g., magnetic disk or tape memory media, over the channel into the main memory of the host. The substantial length of time required to retrieve data from long term storage limits the throughput or usage of the host computer. To minimize this loss of use of the host computer, the host will typically issue a series of requests for data and then perform other tasks while the data is being retrieved from long term disk or tape memory. However, even when this "queueing" is performed there is substantial host computer computation time lost due to the time required for accessing data. Accordingly, it has remained an unsolved problem in the art and it is an object of the present invention to improve host computer throughput by reducing queueing times.
It has been proposed in the prior art that such queueing times be reduced by "staging" data surrounding all data which is the object of a "SEEK" command issued by a host from a disk or tape memory into a solid-state memory of much higher access speed. The solid-state memory is located external to the host, i.e., outboard of the channel from the host. Thus when the host issues subsequent "READ" commands, the data sought may be already contained in the high speed solid-state memory and can be supplied to the host more or less instantaneously. However, if all data sets surrounding records accessed by the host are read into a solid-state memory external to the host as described above, the problem of queueing is not entirely eliminated, as then the channel and director usage time consumed while data is read into cache memory is added to the actual latency time required for the data set to be read from disk into the cache memory.
Moreover, it will be appreciated that there are generally two ways in which data is accessed by a host computer. All the data in a given data set may be called for by the host at a given time, or the host may initiate a separate call for each portion of the data set as each is required. In the first case, addition of the cache memory to the system adds no performance improvement, as but a single latency time is required to satisfy each input/output request. In the second case, wherein each individual host instruction is part of a sequence of instructions typically directed to access successive portions of a physical record such as a tape or disk drive, there is latency time for each call to access a portion of the data set. In this situation, the latency time can be reduced to that of a single access operation involving a single latency time, if successive portions of the data are read into the cache, as the subsequent requests can then be satisfied directly from solid-state memory without involving second and successive physical access operations. That is, if the data is cached in anticipation of a subsequent "SEEK" command it will be available immediately upon issuance of a "READ" command. Accordingly, it is desirable that means be provided for determining which data supplied to a host in response to input/output requests are likely to be part of a sequence of such requests such that only the sequential data called for in requests will be cached. In this way use of the cache will be optimized at no cost to non-sequential operations.
It would not, of course, be impossible for the host computer to issue a signal indicating whether or not a particular data set called for is part of a sequence of such sets. This would of course simplify the decision as to whether or not to "stage" the subsequent record from the long-term data storage means into a cache memory. However, present day computing systems of commercial importance (such as the IBM Corporation's line of computers) do not provide such a signal. Nor is it desirable to modify these computers, nor in particular their operating systems, in order to provide such a signal as such modifications are difficult to implement correctly and are not popular with computer users.
Accordingly, it is desirable to render the caching of data function more efficient by using improved means and methods to determine whether a particular data set sought by a host computer is part of a sequence of such sets (in which event the subsequent sets would be cached) while data which is not amenable to efficient caching is processed in the same manner as in the prior art.
It is a further object of the invention to provide a system in which sequential data sets can be cached so as to improve throughput of a host computer system, without requiring modification to the host.
Yet another object of the invention is to provide a means and method for detecting whether or not a data record is part of a sequence of such records, wherein the means and method operates using information contained within the data record itself, whereby implementation of the method of the invention is rendered simple and relatively inexpensive.